Apple Surface Pesticide Residue Detection Method Based On Hyperspectral Imaging

In order to study the rapid and effective non-destructive detection method of pesticide residues on apple surface, this paper uses hyperspectral imaging technology to verify the feasibility of pesticide residue detection on apple surface. 225 apple samples from two groups were collected to construct the discriminant models of two pesticide residues, i.e., chlorpyrifos and carbendazim. The Hough circle transformation technique was used to determine the Region of Interest (ROI) automatically, and the averaged spectral value of the ROI is calculated as the representative spectrum of the sample. Then the Savitzky-Golay smoothing method was used for spectral denoising. Finally, the discriminant modeling is performed on the whole band with five methods: linear discriminant analysis, linear support vector machine, K nearest neighbor, decision tree and subspace discriminant ensemble. Furthermore, feature band selection was carried out by the successive projection algorithm and subspace discriminant ensemble method, then discriminant models were constructed on the feature band using linear discriminant analysis, linear support vector machine and K nearest neighbor. The experimental results show that the classification accuracy in both the whole band and the selected feature band for the detection of pesticide residues can be up to 95%. For the prediction of pesticide residue concentration, the subspace discriminant ensemble method based on the full band performs better, in which chlorpyrifos pesticide concentration prediction accuracy of up to 95%. The results confirmed the feasibility and effectiveness of hyperspectral imaging to detect pesticide residues on apple surface.